hpc: the first and last cut are the rolling floor's change indicator

[AdaWorldAPI]

Follow-up to #328.

What changes

Drift is now read at the bucket boundaries. FIRST_CUT (1σ) and LAST_CUT (3σ) are located through the active shape twice: once at the anchor, once at the current coordinates.

• Equal motion of the two cuts means translation.
• Unequal motion means dilation or narrowing.
• CutDrift carries both motions and the (Δμ, Δσ) they decompose into:
  • Δσ = (ΔT_first − ΔT_last)·σ / (off_first − off_last)
  • Δμ = ΔT_first − off_first·Δσ/σ
  • On the Gaussian lattice these reduce to Δσ = (ΔT_first − ΔT_last)/2 and Δμ = ΔT_first + Δσ, exactly.
• cut_drift() exposes the indicator on demand.

Cuts decide WHETHER, moments decide WHERE

• MomentsU32 owns the facts, and the first and last cut own the drift decision. Recalibration adopts the exact current facts.
• A checkpoint raises a shift when the cut-implied |Δμ| > σ/2 or |Δσ| > σ/4 of the anchor.
• The FloorShift it raises carries the exact running coordinates (run_mu, run_sigma), never anchor + cut-implied Δ. recalibrate() anchors on those.
• CutDrift.d_mu / d_sigma are diagnostics:
  • Gaussian: the exact parameter decomposition.
  • Empirical: a cut-implied, effective decomposition, quantized by the reservoir's rank resolution. It is not the actual statistical parameter change.
  • Collapsed cuts: when both cuts sit at one offset, spread motion is unobservable by this ruler, so the effective d_sigma is 0.

Behaviour

• Gaussian path: the drift decisions match the previous running-vs-anchor parameter test, and the shift coordinates are unchanged.
• Empirical path: motion is judged in the learned geometry. A dilation that the two cuts cannot express moves no bucket boundary, so it is not drift.
• Resolution: the cut-based Δσ is only as sharp as the gap between the two cuts' offsets. In the test's bimodal sample an 80-unit dilation reads as 84. That is fine for the indicator, and it no longer reaches the anchor.

Tests

• 31/31 pass locally (debug x86-64).
• cuts_decide_whether_moments_decide_where pins the separation. With an empirical shape at an anchor of σ=316 and an actual +80 dilation:
  1. the ruler reads (Δμ, Δσ) = (6, 84);
  2. that reading crosses the σ/4 = 79 criterion;
  3. FloorShift equals the exact (run_mu, run_sigma), not (μ+6, σ+84);
  4. after recalibrate(), the anchor equals the exact MomentsU32 coordinates.
• checkpoint_drift_is_what_the_cuts_see now expects the exact σ of a constant stream (0, floored to 1), not the anchor's 8.
• Disable run: restoring the anchor + Δ construction turns both of those tests red.
• The v3/v4/aarch64 runs, the no_std build and the bench were not run locally, so CI covers them.

🤖 Generated with Claude Code

https://claude.ai/code/session_019HnekoM1EidTwQLS3oFVFm

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

Note

Currently processing new changes in this PR. This may take a few minutes, please wait...

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Essentials

Run ID: e154cd66-8ae4-43c7-a0dd-f3f2ec98925d

📥 Commits

Reviewing files that changed from the base of the PR and between 07c7b4b and 96f1132.

📒 Files selected for processing (1)

• src/hpc/rolling_floor.rs

 _____________________________________
< No code too small, no bug too tall. >
 -------------------------------------
  \
   \   (\__/)
       (•ㅅ•)
       / 　 づ

✨ Finishing Touches

📝 Generate docstrings

• Commit to this branch
• Create a new PR

---

_{Comment @coderabbitai help to get the list of available commands.}

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_cf964493-1e92-4f4a-bf72-10176cd53263)

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_9edd6d4e-01aa-4cf0-952a-305cda8b9ed9)